1) Opening a data file

To open a data file, click on the "Open Data" button and
select a file using the file selection box.  Data files
should be formatted as text files consisting of point
coordinates, separated by spaces.

2) Updating settings

To use the currently displayed settings, click the
"Update Settings" button.  If you change settings and
do not click "Update Settings," the old settings will
remain in effect.

3) Console settings

Range of z0:

The range of z0 should be set to include the probable
range of values for the quantity of interest (z0).
The program only considers points in this range.  If
predictions overflow this range (for instance, if a 95%
prediction interval is not contained in the specified
range), they will be truncated.

The program normalizes data using the Box-Cox family of
transformations for positive data.  If the range of z0
is allowed to extend into the negative numbers, or if z0
is allowed to be too close to zero, the program will
probably report an error.

Mesh size:

The mesh size is the number of points to use in the
discretization of the effective range of z0.  Plots
of densities will have this number of points in them.

The mesh size also governs the precision to which the
program attempts to determine predictions.  For
instance, if the default value of 1000 is used for
an interval from 0.1 to 100, the program will try to
determine the prediction to a precision of 0.1.
Note that increasing the precision will not necessarily
make answers more accurate.

Sample size:

The btg program uses a Monte Carlo integration algorithm
to integrate out transformation and correlation parameters
when it approximates the predictive density function.  See
the manual for more information on this algorithm.

The sample size setting controls the size of the sample
used by this Monte Carlo integration algorithm.  Larger
sample sizes decrease approximation error introduced in
the Monte Carlo integration, but they increase the time
and memory needed to do predictions.

Lambda range:

The parameter lambda governs which member of the Box-Cox
transformation will be used to normalize data.  In the btg
model, the parameter lambda is assumed to be random.  See
the manual for more information.

The btg model uses a uniform prior distribution for lambda.
The lambda range setting controls the support for this
distribution.  The default range from -3 to 3 includes the
values corresponding to logarithmic transformation
(lambda = 0) and to no transformation (lambda = 1).

A single transformation can be chosen by restricting the
lambda range to a single value.  For instance, if the
lambda range is chosen to be from 0 to 0, the logarithmic
transformation will always be used.

Correlation:

The correlation settings governs which family of isotropic
correlation functions should be used.  See the paper for
details.

Trend order:

The program currently assumes that the expected value of
the underlying Gaussian field is polynomial in the point
coordinates.  The trend order setting governs the order of
this underlying polynomial.

The default trend order of zero corresponds to a field with
constant mean.

4) File menu

Load script:
Save script:

The load/save script commands allow settings used in one
session to be saved for future use.  To save the currently
displayed settings, select the save command and enter a file
name.  The settings can be reloaded using the load command.

When settings are saved, the currently displayed settings
will be written out, even if the settings that are being used
have not been updated (see the section on the "Update settings"
button).  Settings are automatically updated when they are
loaded from file.

Quit:

Exits the program.

5) Compute menu

Plot density:

The plot density command writes a plot of the predictive
density p(z0 | z) at a point to disk.  If gnuplot is available,
the output graph is displayed.

Predict:

The predict command computes the prediction at a point and
the 95% prediction interval.

Phi:

The phi command estimates the maximum standard Monte Carlo
error in the computation of p(z0 | z) at an indicated point.

Estimate parameters:

The estimate parameter command computes the maximum
a posteriori parameter estimates for the transformation
parameter lambda and the correlation parameters.  See the
manual for more information on the parameterizations of the
transformation and correlation functions.

Cross-validate:

The cross-validate command performs single-point deletion
cross-validation for each point in the data set.  Output is
written to a file.  When the cross-validation is done, the
output file is displayed in a pop-up browser window.

Map:

The map command maps the prediction and uncertainty values
at each point on a grid over a rectangular region.  The
results are written to file as matrices, with the top row
corresponding to the bottom of the grid, and the first column
corresponding to the left side of the grid.